Sound_Card
Regular
Ya, if AA performance is improved especially on RV770, I highly doubt it my self. Perhaps in certain situations maybe.
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
Ya, if AA performance is improved especially on RV770, I highly doubt it my self. Perhaps in certain situations maybe.
Sound_Card
Regular
No 55nm love for the RV740 or RV710?
Sound_Card
Regular
And??? Your suggesting the performance level should remain the same?
They offer crap performance, although I'm willing to let RV620 slide in that department, I'm not for the all important mid range segment. I'm sure ATi would rather sell mid range cards at the price levels RV670 is selling for and where they use to sell mid range cards for. I really don't see the point on holding out RV740 for 40nm by spring time. Seems way too long and the market demands something more. If it's the case, then Nvidia is going to have a parade with the 9600GT.
I don't see the problem with RV620 and RV635; they are perfectly competitive with G86/G98 and G84/G96 AFAICT and are about what you'd expect them to be given their die size. If we stay on 55nm, either you have substantial architectural performance improvements or it's not worth making new chips... All I'm saying is I doubt RV7xx performance improvements per-mm² are anything but incremental. But I'd love to be wrong, of course! As for RV670, I guess AA improvements for example would be more valuable there...
What's wrong with the 3870 filling that slot? AMD obviously has no problem using RV670 on very cheap parts (3830).
A 50% cut in the GPU core memory bus width means lots of wasted silicon in each die, even if the ALU's remain intact.
Not the best definition of cost effective, especially when you're talking about SKU's in the 79 to 149 dollar slot, with much lower profit margins than the 149 to 249 segment.
Otherwise, there would be a 128bit HD38xx much sooner to combat both the 8800 GS and the 9600 GT (the HD3690 doesn't "count" because it wasn't widely available, at least not in the western markets).
Sound_Card
Regular
I still don't see the point as I don't expect RV740 to be anything less than half of RV770. With that said, a 256bit RV740 should considerably be more attractive than a 128bit RV670(and cost effective) and much more fierce next to a 9600GT as far as performance and margins go.
How about memory chip and PCB layer/complexity costs ?
Sound_Card
Regular
Cost effective and performance effective compared to the competition. Comparing RV670 to a 80nm chip of last gen...? I think the 65nm G94 is what we should be looking at and even though it's bigger, all indication points to better pricing flexibility and I would imagine it being much worse when it goes to 55nm. The 128bit RV670 although cheaper than G94(only in China and nothing official about it coming to the west), gets spanked by it. Not a position I would want to be in if I was ATi.
Instead of waiting on die shrinks long periods away with incremental performance improvements, they need something that scales more effectively from the top that offers better bang. Other wise they will continue to suffer from Nvidia's pricing pressure.
Last edited by a moderator:
I wasn't implying that AMD use the 3830 to fight the 9600GT. They can simply lower the 3870 price some more. Isn't RV670 still a fair bit smaller than G94? We can expect Nvidia to keep the 9600GT around for a long time so I don't see the urgency in replacing the 3870. Besides, for all we know RV740 is slower than RV670....
As Arun said, AMD already has offerings competitive with the 8400/9300/9400 and 8600 lines. The 9500GT should slot in somewhere between the 3650 and 3850 cards.
Berek
Regular
I noticed your projections illustrated the X2 coming out in Q4... if that is true, I would probably get to single cards and SLI them together, as I want to build a system right away upon my return home in July.
My question: Is there any differences between an X2 and two SLI'd/Crossfire cards? I see these differences at this time:
1) Two separate cards might draw more power
2) One X2 card costs slightly more than two separate
3) Performance differences are negligible, if any between the two setups.
4) SLI'd/Crossfire cards are naturally more inconvenient on the whole vs. one card.
Anything I'm missing or got wrong there?
How would a 4 SIMD, 32 TMU setup work? You'd have six quads per SIMD but 8 quad TMU's.....
Sound_Card
Regular
If AA performance is improved in the HD4xxx architecture, then I would not be too sure about that. I'm led to believe that RV770(256bit) consist of 2 RBE's, 4 TEU's, and 48sp's. Something like that I would imagine being slightly smaller than RV670. A much more appealing offering than the 128bit RV670 and stronger competitor to the 8800gs and possibly even the 9600GT. And where is the HD 3870 going to be when RV770 pro models hit?
http://www.fudzilla.com/index.php?option=com_content&task=view&id=5681&Itemid=1
I hope Fudo is right.
Yeah I'm not saying anything different. I'm just saying that right now Nvidia doesn't have anything in the mid/low segments that AMD cannot adequately counter with existing solutions.
On another note: if RV770 does indeed have 480 ALU's it looks like it will fall short of the 1TFlop mark. I'm not really confident that they can hit the required 1Ghz+ core speed. Nvidia probably doesnt have a chance of hitting 1Tflop of MADDs either.
On another note: if RV770 does indeed have 480 ALU's it looks like it will fall short of the 1TFlop mark. I'm not really confident that they can hit the required 1Ghz+ core speed. Nvidia probably doesnt have a chance of hitting 1Tflop of MADDs either.
Jawed
Legend
4 or 5%?
Look at the transistor counts for R600 versus RV670 and note that the latter has 50% of the former's MCs.
Jawed
Jawed
Legend
The same way that 3 quads per SIMD worked with 4 quad TMUs in R580 (1 quad TMU per SIMD).
Jawed
And then add trannies for DX10.1 and UVD...
- Status
Similar threads
